Debian Codegraph @

upstream/debian-unstable/man2/process_madvise.2

raw
1 .\" Copyright 2021, Suren Baghdasaryan <[email protected]>
2 .\" Copyright 2021, Minchan Kim <[email protected]>
3 .\" Copyright, the authors of the Linux man-pages project
4 .\"
5 .\" SPDX-License-Identifier: Linux-man-pages-copyleft
6 .\"
7 .\" Commit ecb8ac8b1f146915aa6b96449b66dd48984caacc
8 .\"
9 .TH process_madvise 2 2026-02-22 "Linux man-pages 6.18"
10 .SH NAME
11 process_madvise \- give advice about use of memory to a process
12 .SH LIBRARY
13 Standard C library
14 .RI ( libc ,\~ \-lc )
15 .SH SYNOPSIS
16 .nf
17 .B #include <sys/mman.h>
18 .P
19 .BI "ssize_t process_madvise(size_t " n ;
20 .BI " int " pidfd ", const struct iovec " iovec [ n ],
21 .BI " size_t " n ", int " advice \
22 ", unsigned int " flags );
23 .fi
24 .SH DESCRIPTION
25 The
26 .BR process_madvise ()
27 system call is used to give advice or directions to the kernel about the
28 address ranges of another process or of the calling process.
29 It provides the advice for the address ranges described by
30 .I iovec
31 and
32 .IR n .
33 The goal of such advice is to improve system or application performance.
34 .P
35 The
36 .I pidfd
37 argument is a PID file descriptor (see
38 .BR pidfd_open (2))
39 that specifies the process to which the advice is to be applied.
40 .P
41 The pointer
42 .I iovec
43 points to an array of
44 .I iovec
45 structures, described in
46 .BR iovec (3type).
47 .P
48 .I n
49 specifies the number of elements in the array of
50 .I iovec
51 structures.
52 This value must be less than or equal to
53 .B IOV_MAX
54 (defined in
55 .I <limits.h>
56 or accessible via the call
57 .IR sysconf(_SC_IOV_MAX) ).
58 .P
59 If manipulating another process,
60 or before Linux 6.13,
61 the
62 .I advice
63 argument is one of the following values:
64 .TP
65 .B MADV_COLD
66 See
67 .BR madvise (2).
68 .TP
69 .B MADV_COLLAPSE
70 See
71 .BR madvise (2).
72 .TP
73 .B MADV_PAGEOUT
74 See
75 .BR madvise (2).
76 .TP
77 .B MADV_WILLNEED
78 See
79 .BR madvise (2).
80 .P
81 Since Linux 6.13,
82 when manipulating the calling process,
83 any advice flag is permitted.
84 .P
85 The
86 .I flags
87 argument is reserved for future use;
88 currently,
89 this argument must be specified as 0.
90 .P
91 The
92 .I n
93 and
94 .I iovec
95 arguments are checked before applying any advice.
96 If
97 .I n
98 is too big, or
99 .I iovec
100 is invalid,
101 then an error will be returned immediately and no advice will be applied.
102 .P
103 The advice might be applied to only a part of
104 .I iovec
105 if one of its elements points to an invalid memory region in the
106 remote process.
107 No further elements will be processed beyond that point.
108 (See the discussion regarding partial advice in RETURN VALUE.)
109 .P
110 .\" commit 96cfe2c0fd23ea7c2368d14f769d287e7ae1082e
111 Since Linux 5.12,
112 permission to apply advice to another process is governed by
113 ptrace access mode
114 .B PTRACE_MODE_READ_FSCREDS
115 check (see
116 .BR ptrace (2));
117 in addition,
118 because of the performance implications of applying the advice,
119 the caller must have the
120 .B CAP_SYS_NICE
121 capability
122 (see
123 .BR capabilities (7)).
124 .SH RETURN VALUE
125 On success,
126 .BR process_madvise ()
127 returns the number of bytes advised.
128 This return value may be less than the total number of requested bytes,
129 if an error occurred after some
130 .I iovec
131 elements were already processed.
132 The caller should check the return value to determine whether a partial
133 advice occurred.
134 .P
135 On error, \-1 is returned and
136 .I errno
137 is set to indicate the error.
138 .SH ERRORS
139 .TP
140 .B EBADF
141 .I pidfd
142 is not a valid PID file descriptor.
143 .TP
144 .B EFAULT
145 The memory described by
146 .I iovec
147 is outside the accessible address space of the process referred to by
148 .IR pidfd .
149 .TP
150 .B EINVAL
151 .I flags
152 is not 0.
153 .TP
154 .B EINVAL
155 The sum of the
156 .I iov_len
157 values of
158 .I iovec
159 overflows a
160 .I ssize_t
161 value.
162 .TP
163 .B EINVAL
164 .I n
165 is too large.
166 .TP
167 .B ENOMEM
168 Could not allocate memory for internal copies of the
169 .I iovec
170 structures.
171 .TP
172 .B EPERM
173 The caller does not have permission to access the address space of the process
174 .IR pidfd .
175 .TP
176 .B ESRCH
177 The target process does not exist (i.e., it has terminated and been waited on).
178 .P
179 See
180 .BR madvise (2)
181 for
182 .IR advice -specific
183 errors.
184 .SH STANDARDS
185 Linux.
186 .SH HISTORY
187 Linux 5.10,
188 .\" Linux commit ecb8ac8b1f146915aa6b96449b66dd48984caacc
189 glibc 2.36.
190 .\" glibc commit d19ee3473d68ca0e794f3a8b7677a0983ae1342e
191 .P
192 Support for this system call is optional,
193 depending on the setting of the
194 .B CONFIG_ADVISE_SYSCALLS
195 configuration option.
196 .P
197 When this system call first appeared in Linux 5.10,
198 permission to apply advice to another process was entirely governed by
199 ptrace access mode
200 .B PTRACE_MODE_ATTACH_FSCREDS
201 check (see
202 .BR ptrace (2)).
203 This requirement was relaxed in Linux 5.12 so that the caller didn't require
204 full control over the target process.
205 .SH SEE ALSO
206 .BR madvise (2),
207 .BR pidfd_open (2),
208 .BR process_vm_readv (2),
209 .BR process_vm_write (2)